Structural health monitoring (SHM) of indoor environments is essential for early crack detection, signaling potential structural failures. Manual inspections are labor-intensive, error-prone, and hazardous, driving the need for automation. This paper introduces a compact end-to-end autonomous robotic system for indoor crack monitoring, integrating online mapping, waypoint navigation, vision-based detection, large language model (LLM)-assisted screening, coordinate transformation, and evidence collection. The system uses a mobile platform to create a 2D occupancy grid map via simultaneous localization and mapping (SLAM). At predefined waypoints, a fine-tuned YOLOv5 detector identifies cracks from camera feeds, with outputs screened by a lightweight LLM to minimize false positives through contextual reasoning. Verified coordinates are transformed from camera to global map using time-stamped transformations. For multi-view coverage, the robot executes two 90° counterclockwise rotations per waypoint. It then navigates to cracks for high-resolution close-up imaging. In the real case study, YOLOv5 shows stable loss curves, confusion matrix with precision (0.8707) and recall (0.7840), and mAP@0.5 of 0.8262 at optimal epoch. Results are visualized on a map overlaying crack positions and evidence images for intuitive review. This enhances SHM accuracy, efficiency, and safety, with broader infrastructure applications.

Crack,Robot,Indoor Positioning,Monitoring,LLM,Autonomous Vehicle